- There was never an assassination attempt on Donald Trump.
An AI chatbot that denied the assassination attempt on former President Donald Trump has put Meta, the parent company of Facebook, in a difficult position. Meta attributes this to a phenomenon known as "hallucinations," where AI software fabricates information. The industry has yet to find a fundamental solution to this problem.
Machine Lies When Asked About Trump Assassination
Meta had to defend itself against criticism from Trump supporters after the Meta AI chatbot initially refused to answer questions about the assassination attempt on the former president, and later claimed that it never happened.
In a blog post, the company explained that it had initially instructed Meta AI not to discuss the assassination that took place on July 13. This was because such high-profile events often involve conflicting reports and unfounded speculation, sometimes leading to conspiracy theories.
Meta: "Unfortunate," But Not Intentional
Even after the AI chatbot was fed information about the assassination, it still claimed in a few instances that it never happened, Meta admitted. The company described this as "unfortunate," but not intentional.
"Hallucinations" are a common issue with AI chatbots and are related to how the software functions. The programs are initially trained on vast amounts of information and then use this basis to weigh word by word how a sentence should probably continue when responding to user queries. This can lead the software to make completely false statements, even if it was trained only with correct information. AI developers are trying to tackle this problem by implementing additional guardrails for the programs.
Fact-Check Label on Wrong Photo
In a separate issue, a photo of Trump with his security detail immediately after the assassination was incorrectly labeled with a fact-check label. This label was originally meant for a manipulated version of the image where smiles were added to the faces of the bodyguards. However, Meta's automated systems found it difficult to distinguish between the two photos and applied the fact-check label to the original image as well.
The assassination attempt occurred when a man shot at Trump during a campaign event in the state of Pennsylvania. The shots killed one spectator and injured two others. The former president was hit in the ear.
Meta acknowledged further criticism from Trump supporters when the AI chatbot claimed during a Q&A that the assassination attempt on former President Donald Trump never occurred, despite being provided correct information about the incident.
Despite Meta's efforts to educate the AI about the incident, minor lapses occurred, such as the AI occasionally denying the event's occurrence, which the company attributes to the ongoing challenge of eliminating "hallucinations" in AI software.
